Abstract

Locally adapted parameterizations of a model (such as locally weighted regression) are expressive but often suffer from high variance. We describe an approach for reducing this variance, based on the idea of estimating simultaneously a transformed space for the model and locally adapted parameterizations expressed in the new space. We present a new problem formulation that captures this idea and illustrate it in the im portant context of time varying models. We develop an algorithm for learning a set of bases for approximating a time varying sparse network; each learned basis constitutes an archetypal sparse network structure. We also provide an extension for learning task-specific bases. We present empirical results on synthetic data sets, as well as on a BCI EEG classification task.